The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses various command-line utilities to inject system failures, including 'aws-cli' for terminating EC2 instances and rebooting RDS databases, 'kubectl' for managing Kubernetes resources, and 'toxiproxy-cli' for network manipulation.
Evidence includes a bash script in references/chaos-tools.md that terminates instances based on an Auto Scaling Group query.
Evidence in references/infrastructure-chaos.md shows usage of stress-ng for resource exhaustion and pumba for container-level chaos.
[EXTERNAL_DOWNLOADS]: The skill integrates with external chaos engineering frameworks and monitoring tools.
references/chaos-tools.md contains a GitHub Action workflow that downloads the Litmus operator manifest directly from litmuschaos.github.io using kubectl apply.
The skill also references official repositories for toxiproxy and stress-ng for local installation via brew and apt-get.
[PRIVILEGE_ESCALATION]: Several scripts provided in the skill utilize high-privilege execution to modify system-level configurations.
references/infrastructure-chaos.md includes a DNSChaos class that uses sudo to append entries to /etc/hosts for simulating DNS failures.
The resource exhaustion section in references/infrastructure-chaos.md uses sudo apt-get to install the stress-ng utility.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface where untrusted configuration data or tool outputs could influence agent actions.
Ingestion points: Reads experiment manifests (YAML) and monitoring metrics (JSON/Prometheus) to determine experiment status and recovery success.
Boundary markers: None identified in the prompt templates or processing scripts.
Capability inventory: Extensive capabilities including cloud infrastructure modification (aws), cluster modification (kubectl), and local system modification (sudo).
Sanitization: No explicit sanitization of metric queries or resource labels is shown before they are interpolated into command strings.

chaos-engineer